Systematic review and meta-analysis of depression, anxiety, and suicidal ideation among Ph.D. students

University administrators and mental health clinicians have raised concerns about depression and anxiety among Ph.D. students, yet no study has systematically synthesized the available evidence in this area. After searching the literature for studies reporting on depression, anxiety, and/or suicidal ideation among Ph.D. students, we included 32 articles. Among 16 studies reporting the prevalence of clinically significant symptoms of depression across 23,469 Ph.D. students, the pooled estimate of the proportion of students with depression was 0.24 (95% confidence interval [CI], 0.18–0.31; I2 = 98.75%). In a meta-analysis of the nine studies reporting the prevalence of clinically significant symptoms of anxiety across 15,626 students, the estimated proportion of students with anxiety was 0.17 (95% CI, 0.12–0.23; I2 = 98.05%). We conclude that depression and anxiety are highly prevalent among Ph.D. students. Data limitations precluded our ability to obtain a pooled estimate of suicidal ideation prevalence. Programs that systematically monitor and promote the mental health of Ph.D. students are urgently needed.

in coursework 59 . Other studies identified a higher prevalence of mental ill-health among women 54 ; lesbian, gay, bisexual, transgender, and queer (LGBTQ) students 42,54,60 ; and students with multiple intersecting identities 54 . Several studies identified correlates of mental health problems including: project-and supervisor-related issues, stress about productivity, and self-doubt 53,62 ; uncertain career prospects, poor living conditions, financial stressors, lack of sleep, feeling devalued, social isolation, and advisor relationships 61 ; financial challenges 38 ; difficulties with work-life balance 58 ; and feelings of isolation and loneliness 52 . Despite these challenges, help-seeking appeared to be limited, with only about one-quarter of Ph.D. students reporting mental health problems also reporting that they were receiving treatment 40,52 . Risk of bias. Twenty-one of 32 articles were assessed as having low risk of bias (Supplementary Table S2).
Five articles received one point for all five categories on the risk of bias assessment (lowest risk of bias), and one article received no points (highest risk). The mean risk of bias score was 3.22 (standard deviation, 1.34; median, 4; IQR, [2][3][4]. Restricting the estimation sample to 12 studies assessed as having low risk of bias, the estimated proportion of Ph.D. students with depression was 0.25 (95% CI, 0.18-0.33; 95% PI, 0.04-0.57; I 2 = 99.11%), nearly identical to the primary estimate, with no reduction in heterogeneity. The estimated proportion of Ph.D. students with anxiety, among the 7 studies assessed as having low risk of bias, was 0.12 (95% CI, 0.07-0.17; 95% PI, 0.01-0.34; I 2 = 98.17%), again with no appreciable reduction in heterogeneity.

Discussion
In our meta-analysis of 16 studies representing 23,469 Ph.D. students, we estimated that the pooled prevalence of clinically significant symptoms of depression was 24%. This estimate is consistent with estimated prevalence rates in other high-stress biomedical trainee populations, including medical students (27%) 30 , resident physicians (29%) 65 , and postdoctoral research fellows (29%) 66 . In the sample of nine studies representing 15,626 Ph.D. students, we estimated that the pooled prevalence of clinically significant symptoms of anxiety was 17%. While validated screening instruments tend to over-identify cases of depression (relative to structured clinical interviews) by approximately a factor of two 67,68 , our findings nonetheless point to a major public health problem among Ph.D. students. Available data suggest that the prevalence of depressive and anxiety disorders in the general population ranges from 5 to 7% worldwide 69,70 . In contrast, prevalence estimates of major depressive   Further underscoring the importance of this public health issue, Ph.D. students face unique stressors and uncertainties that may put them at increased risk for mental health and substance use problems. Students grapple with competing responsibilities, including coursework, teaching, and research, while also managing interpersonal relationships, social isolation, caregiving, and financial insecurity 3,10 . Increasing enrollment in doctoral degree programs has not been matched with a commensurate increase in tenure-track academic job opportunities, intensifying competition and pressure to find employment post-graduation 5 . Advisor-student power relations rarely offer options for recourse if and when such relationships become strained, particularly in the setting of sexual harassment, unwanted sexual attention, sexual coercion, and rape [74][75][76][77][78] . All of these stressors may be magnified-and compounded by stressors unrelated to graduate school-for subgroups of students who are underrepresented in doctoral degree programs and among whom mental health problems are either more prevalent and/or undertreated compared with the general population, including Black, indigenous, and other people of color 13,79,80 ; women 81,82 ; first-generation students 14,15 ; people who identify as LGBTQ [83][84][85] ; people with disabilities; and people with multiple intersecting identities. www.nature.com/scientificreports/ Structural-and individual-level interventions will be needed to reduce the burden of mental ill-health among Ph.D. students worldwide 31,86 . Despite the high prevalence of mental health and substance use problems 87 , Ph.D. students demonstrate low rates of help-seeking 40,52,88 . Common barriers to help-seeking include fears of harming one's academic career, financial insecurity, lack of time, and lack of awareness 89-91 , as well as health care systems-related barriers, including insufficient numbers of culturally competent counseling staff, limited access to psychological services beyond time-limited psychotherapies, and lack of programs that address the specific needs either of Ph.D. students in general 92 or of Ph.D. students belonging to marginalized groups 93,94 . Structural interventions focused solely on enhancing student resilience might include programs aimed at reducing stigma, fostering social cohesion, and reducing social isolation, while changing norms around help-seeking behavior 95,96 . However, structural interventions focused on changing stressogenic aspects of the graduate student environment itself are also needed 97 , beyond any enhancements to Ph.D. student resilience, including: undercutting power differentials between graduate students and individual faculty advisors, e.g., by diffusing power among multiple faculty advisors; eliminating racist, sexist, and other discriminatory behaviors by faculty advisors 74,75,98 ; valuing mentorship and other aspects of "invisible work" that are often disproportionately borne by women faculty and faculty of color 99,100 ; and training faculty members to emphasize the dignity of, and adequately prepare Ph.D. students for, non-academic careers 101,102 .
Our findings should be interpreted with several limitations in mind. First, the pooled estimates are characterized by a high degree of heterogeneity, similar to meta-analyses of depression prevalence in other populations 30,65,[103][104][105] . Second, we were only able to aggregate depression prevalence across 16 studies and anxiety prevalence across nine studies (the majority of which were conducted in the U.S.) -far fewer than the 183 studies included in a meta-analysis of depression prevalence among medical students 30 and the 54 studies included in a meta-analysis of resident physicians 65 . These differences underscore the need for more rigorous study in this critical area. Many articles were either excluded from the review or from the meta-analyses for not meeting inclusion criteria or not reporting relevant statistics. Future research in this area should ensure the systematic www.nature.com/scientificreports/ collection of high-quality, clinically relevant data from a comprehensive set of institutions, across disciplines and countries, and disaggregated by graduate student type. As part of conducting research and addressing student mental health and wellbeing, university deans, provosts, and chancellors should partner with national survey and program institutions (e.g., Graduate Student Experience in the Research University [gradSERU] 106 , the American College Health Association National College Health Assessment [ACHA-NCHA], and HealthyMinds). Furthermore, federal agencies that oversee health and higher education should provide resources for these efforts, and accreditation agencies should require monitoring of mental health and programmatic responses to stressors among Ph.D. students. Third, heterogeneity in reporting precluded a meta-analysis of the suicidality outcomes among the few studies that reported such data. While reducing the burden of mental health problems among graduate students is an important public health aim in itself, more research into understanding non-suicidal self-injurious behavior, suicide attempts, and completed suicide among Ph.D. students is warranted. Fourth, it is possible that the grey literature reports included in our meta-analysis are more likely to be undertaken at research-intensive institutions 52,60,61 . However, the direction of bias is unpredictable: mental health problems among Ph.D. students in research-intensive environments may be more prevalent due to detection bias, but such institutions may also have more resources devoted to preventive, screening, or treatment efforts 92 . Fifth, inclusion in this meta-analysis and systematic review was limited to those based on community samples. Inclusion of clinic-based samples, or of studies conducted before or after specific milestones (e.g., the qualifying examination or dissertation prospectus defense), likely would have yielded even higher pooled prevalence estimates of mental health problems. And finally, few studies provided disaggregated data according to sociodemographic factors, stage of training (e.g., first year, pre-prospectus defense, all-but-dissertation), or discipline of study. These factors might be investigated further for differences in mental health outcomes.
Clinically significant symptoms of depression and anxiety are pervasive among graduate students in doctoral degree programs, but these are understudied relative to other trainee populations. Structural and clinical interventions to systematically monitor and promote the mental health and wellbeing of Ph.D. students are urgently needed.

Methods
This systematic review and meta-analysis follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) approach (Supplementary Table S3) 107 . This study was based on data collected from publicly available bibliometric databases and did not require ethical approval from our institutional review boards.
Eligibility criteria. Studies were included if they provided data on either: (a) the number or proportion of Ph.D. students with clinically significant symptoms of depression or anxiety, ascertained using a validated scale; or (b) the mean depression or anxiety symptom severity score and its standard deviation among Ph.D. students. Suicidal ideation was examined as a secondary outcome.
We excluded studies that focused on graduate students in non-doctoral degree programs (e.g., Master of Public Health) or professional degree programs (e.g., Doctor of Medicine, Juris Doctor) because more is known about mental health problems in these populations 30,[108][109][110] and because Ph.D. students face unique uncertainties. To minimize the potential for upward bias in our pooled prevalence estimates, we excluded studies that recruited students from campus counseling centers or other clinic-based settings. Studies that measured affective states, or state anxiety, before or after specific events (e.g., terrorist attacks, qualifying examinations) were also excluded.
If articles described the study sample in general terms (i.e., without clarifying the degree level of the participants), we contacted the authors by email for clarification. Similarly, if articles pooled results across graduate students in doctoral and non-doctoral degree programs (e.g., reporting a single estimate for a mixed sample of graduate students), we contacted the authors by email to request disaggregated data on the subsample of Ph.D. students. If authors did not reply after two contact attempts spaced over 2 months, or were unable to provide these data, we excluded these studies from further consideration.
Search strategy and data extraction. PubMed, Embase, PsycINFO, ERIC, and Business Source Complete were searched from inception of each database to November 5, 2019. The search strategy included terms related to mental health symptoms (e.g., depression, anxiety, suicide), the study population (e.g., graduate, doctoral), and measurement category (e.g., depression, Columbia-Suicide Severity Rating Scale) (Supplementary Table S4). In addition, we searched the reference lists and the grey literature.
After duplicates were removed, we screened the remaining titles and abstracts, followed by a full-text review. We excluded articles following the eligibility criteria listed above (i.e., those that were not focused on Ph.D. students; those that did not assess depression and/or anxiety using a validated screening tool; those that did not report relevant statistics of depression and/or anxiety; and those that recruited students from clinic-based settings). Reasons for exclusion were tracked at each stage. Following selection of included articles, two members of the research team extracted data and conducted risk of bias assessments. Discrepancies were discussed with a third member of the research team. Key extraction variables included: study design, geographic region, sample size, response rate, demographic characteristics of the sample, screening instrument(s) used for assessment, mean depression or anxiety symptom severity score (and its standard deviation), and the number (or proportion) of students experiencing clinically significant symptoms of depression or anxiety.
Risk of bias assessment. Following prior work 30,65 , the Newcastle-Ottawa Scale 111 was adapted and used to assess risk of bias in the included studies. Each study was assessed across 5 categories: sample representativeness, sample size, non-respondents, ascertainment of outcomes, and quality of descriptive statistics reporting www.nature.com/scientificreports/ (Supplementary Information S5). Studies were judged as having either low risk of bias (≥ 3 points) or high risk of bias (< 3 points).

Analysis and synthesis.
Before pooling the estimated prevalence rates across studies, we first transformed the proportions using a variance-stabilizing double arcsine transformation 112 . We then computed pooled estimates of prevalence using a random effects model 113 . Study specific confidence intervals were estimated using the score method 114,115 . We estimated between-study heterogeneity using the I 2 statistic 116 . In an attempt to reduce the extent of heterogeneity, we re-estimated pooled prevalence restricting the analysis to studies conducted in the United States and to studies in which depression assessment was based on the 9-item Patient Health Questionnaire (PHQ-9) 117 . All analyses were conducted using Stata (version 16; StataCorp LP, College Station, Tex.). Where heterogeneity limited our ability to summarize the findings using meta-analysis, we synthesized the data using narrative review. www.nature.com/scientificreports/